Secondary Outcomes


Method and timing of outcome measurement: A systematic method was applied to measure outcomes both method wise and timing wise. Regarding methods, standardized measurement instruments were used for those outcomes which can be measured objectively, such as lipid levels, BMI, and body weight. For other outcome measures such as health-related quality of life and socioeconomic effects, standardized/valid scales of measurements were used when available, or widely acceptable definitions of the outcomes were followed.

All the studies were categorized in two broad groups based on timing of outcome measurement "short-term" group with timing of at least 3 weeks and "long-term" group of more than 6 months. Outcomes including lipid levels, BMI, and body weight were considered in the short-term group whereas other outcomes such as health-related quality of life, adverse events, morbidity, and mortality, along with lipid levels, BMI, and body weight, were considered in the long-term group.

Data collection, analysis, and calculation of treatment effects: We used pre-designed standard data abstraction forms to collect data and software Revman, Version 5.3 (The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark) to enter, analyze, and synthesize the data. Results were presented in forest plots and data summary tables. Dichotomous data were calculated in risk ratio or odds ratio with a 95% confidence interval (CI) whereas the continuous data were calculated in mean difference with a 95% CI and standardized mean difference with a 95% CI.

Meta-analysis method: Inverse variance method was used for meta-analysis. Dichotomous data were expressed as odds ratios (ORs) or risk ratios (RRs) with 95% confidence intervals (CIs). Continuous data were expressed as mean differences (MDs) with 95% CIs. Time-to-event data were expressed as hazard ratios (HRs) with 95% CIs. The formulas for meta-analysis were used as described by Deeks and Higgins in the supplementary statistical guidelines for the software Revman 5.3 [31]. The level at which randomization occurred was closely monitored, such as cross-over trials, cluster-randomized trials, and multiple observations for the same outcome.

#### 2.1.3. Assessment of Risk of Bias in Included Studies

DG and RS assessed risk in individual studies independently. Disagreements were resolved by consensus, or by consultation with SR. The Cochrane Collaboration's tool was used to assess the risk of bias [29,32]. The following criteria were assessed for this purpose:


Risk of bias criteria was judged as "low risk", "high risk", or "unclear risk" and individual bias items were evaluated as described in the *Cochrane Handbook for Systematic Reviews of Interventions* [29].

Missing Data: Missing data were obtained from authors where possible, and reasons for missing data (attrition rates, e.g., drop-outs, losses to follow-up and withdrawals, issues of missing data and imputation methods (e.g., last observation carried forward [LOCF])) were investigated and critically appraised. Missing standard deviations (SD) were imputed (average of SD of studies where reported) and the impact of imputation on meta-analyses was investigated by sensitivity analyses.

Assessment of Heterogeneity: Causes of any significant clinical, methodological, or statistical heterogeneity were explored but the pooled effect estimate in a meta-analysis was still presented. Heterogeneity was identified through visual inspection of the forest plots and by using a standard chi-square test *α* and the I2 statistic < 75% [33]. If 10 or more studies were included investigating a particular outcome, funnel plots were used to assess small study effects. Several explanations can be offered for the asymmetry of a funnel plot, including true heterogeneity of effect with respect to trial size, poor methodological design (and hence bias of small trials), and publication bias. Therefore, results were interpreted carefully [34].

#### 2.1.4. Data Analyses

Revman (Version 5.3) was used to compute effect sizes as well as other statistical information such as *p*-values, t-scores, Q statistics, and confidence intervals. Forest plots, funnel plots, and data summary tables were created utilizing this software. Unless there was good evidence for homogeneous effects across studies, primarily low risk of bias data was summarized using a random-effects model [35]. Random-effects meta-analyses were interpreted with due consideration of the whole distribution of effects by presenting a prediction interval [36]. A prediction interval specifies a predicted range for the true treatment effect in an individual study [37]. Statistical analyses were performed according to the statistical guidelines in the latest version of the *Cochrane Handbook for Systematic Reviews of Interventions* [29].

#### *2.2. Subgroup Analysis and Investigation of Heterogeneity*

The following characteristics were expected to introduce clinical heterogeneity and, when possible, subgroup analyses were conducted:


#### *2.3. Sensitivity Analysis*

Sensitivity analyses was performed in order to explore the influence of the following factors (when applicable) on effect sizes:


• Restricting the analysis to studies using the following filters: diagnostic criteria, imputation, language of publication, source of funding (industry versus other), and country.

The robustness of the results was tested by repeating the analysis using different measures of effect size (RR, odds ratio (OR), etc.) and different statistical models (fixedeffect and random-effects models).

#### *2.4. Including Non-Randomized Studies*

When there were only a small number of randomized studies identified for systematic review and meta-analysis, non-randomized studies were also included. These non-randomized studies may be quasi-randomized, controlled clinical trials, or simply before-after clinical trials.

However, data from both randomized and non-randomized studies were not combined together in the same analysis as this may affect the strength of the evidence. The guidelines from *Cochrane Handbook of Systematic Reviews and Meta-Analysis* says that "where randomized trial evidence is desired but unlikely to be available, eligibility criteria could only be structured to say that nonrandomized studies would only be included where randomized trials are found not to be available. In time, as such a review is updated the non-randomized studies may be dropped when randomized trials become available" [29] (p. 397).

## **3. Results**

A total of 1756 potentially relevant studies were found by searching the databases MEDLINE, CENTRAL, AMED, EMBASE, WHO ICTRP, Dhara online, AYUSH research portal, Clinicaltrials.gov, and INDMED. Through hand searches, 18 more studies were identified. After duplication and screening of the titles of obtained records, a total of 447 studies were considered for further screening. After perusal of the titles and abstracts, 387 studies were excluded due to the following reasons: focusing on herbs not part of Ayurveda (Western and Chinese herbs), reviews, observational studies, and not meeting the inclusion criteria. Sixty studies were found potentially eligible at this stage and the full papers were obtained. Among these, 14 studies were non-randomized, 8 studies did not fulfill initial inclusion criteria, 4 were either incomplete or potentially ongoing, and the full text of 2 studies could not be obtained. Hence, only 32 studies were ultimately included in the systematic review, of which 24 studies were qualified to be included in the meta-analysis. The characteristics of the included studies are included in Table 1. An adapted PRISMA [28] flow-chart of the study selection appears in Figure 1.

Excluded studies: Among potentially relevant studies, 28 were excluded for the following reasons. Two studies were of short duration, fourteen were non-randomized, and two could not be included as their full text could not be retrieved. Ten studies did not meet the minimum inclusion criteria. Studies investigating the use of Western herbal preparations that are not used in Ayurveda and pharmacological studies were excluded. Studies of garlic using garlic oil or aged garlic were also excluded because these are not described in classical Ayurvedic literature.


**Table 1.** Characteristics of included studies.

#### *Medicina* **2021**, *57*, 546


**Table 1.** *Cont*.



**Table 1.** *Cont*.

**Figure 1.** Study flow diagram on Ayurvedic herbal preparations for hypercholesterolemia.

Risk of bias assessment: Although the majority of the studies were randomized and double-blind, they failed to provide details of random sequence generation and allocation concealment. It was definitely inadequate in the majority of the studies. Only six studies out of 32 explained the random sequence generation. Two studies had high risk of bias in blinding whereas seven studies did not specify their blinding status. Both participants and investigators were blinded in 22 studies. In general, the blinding was achieved by using identical looking treatment and placebo tablets or capsules. However, the blinding of participants and investigators was more common than the blinding of outcome assessors. There was less information about the outcome assessment methods and personnel. Fifteen studies were prone to have attrition bias as the dropouts were not included in the final analysis. In one of the studies [59], 50% withdrawal was reported and the explanation was given as "unavoidable circumstances". In another study [48], 39 out of 64 participants in the treatment group and 34 out of 59 participants in the control group completed the study. Since many of the studies did not provide information on their protocol, it is hard to say if selective outcome reporting bias existed. However, based on the methods in the studies, all but three of the studies did not seem to have selective outcome reporting bias. A table of risk of bias assessment is given in Table 2.




**Table 2.** *Cont*.

L: low risk; H: high risk; U: unclear risk.

#### *3.1. Effects of Ayurvedic Herbs*

#### 3.1.1. Total Cholesterol (TC) (mg/dL)

Overall, Ayurvedic herbal formulations were found to be effective in reducing total cholesterol by approximately 7.5%. A meta-analysis of 24 randomized controlled trials on four different Ayurvedic interventions, namely garlic, guggulu, *Nigella sativa,* and a combination of garlic and guggulu (*Lashunadi Guggulu*), involved a total of 1386 participants with 699 in the Ayurvedic group and 687 participants in the control group (Table 3).

Figure 2, a forest plot of the meta-analysis, shows that the most effective intervention is Lashunadi Guggulu (garlic + guggulu), with a reduction of 38.28 mg/dL in TC (95% C.I.: −55.11 to 21.14; *p* < 0.00001). The second most effective intervention was guggulu (*Commiphora mukul*), reducing TC by 16.78 mg/dL (95% C.I.: −13.96 to −2.61; *p* = 0.02) or almost 8.5% of borderline high TC levels. The third most effective intervention for reducing high TC was found to be garlic. Analysis of findings from 11 studies comparing

404 participants taking garlic with 409 participants on a placebo showed that garlic reduces TC by 12.45 mg/dL (95% C.I.: −18.68 to −6.22, *p* < 00001). Finally, the intervention with the least effect was found to be *Nigella sativa,* reducing TC by 9.28 mg/dL (95% C.I.: −17.36 to −1.19; *p* = 0.02).

**Table 3.** Effect of Ayurvedic herbal preparations in total cholesterol (mg/dL).



**Figure 2.** Forest plot on effect of Ayurvedic herbal preparations on total cholesterol (mg/dL).

#### 3.1.2. LDL Cholesterol (mg/dL)

As shown in Table 4 and Figure 3, garlic was found to reduce LDL-C by 10.37 mg/dL (95% C.I.: −17.58 to −3.16; *p* = 0.005). This result is nearly 8% of the borderline LDL-C levels. The heterogeneity between garlic studies was 66%. As compared to the placebo, guggulu was found to reduce LDL-C by −18.78 mg/dL (95% C.I.: −34.07 to −3.48; *p* = 0.02). Unlike the results for total cholesterol, *Nigella sativa* did not have significant effects on a reduction in LDL-C (2.12 mg/dL (95% C.I.: −7.85 to 3.6; *p* = 0.47)), as shown in Figure 3. Altogether, these studies included 163 participants, of which 84 people were in an intervention group and the remaining 79 were in the control group.

**Table 4.** Effect of Ayurvedic herbal preparations on LDL-C (mg/dL).



**Figure 3.** Forest plot on the effect of Ayurvedic herbal preparations on LDL-C (mg/dL).

#### 3.1.3. Triglycerides (mg/dL)

Meta-analyses of the four interventions showed (Table 5 and Figure 4) garlic to be the least effective in reducing raised TG levels (3.1 mg/dL (95% C.I.: −16.63 to 10.42; *p* value = 0.65)), as shown in Figure 4. *N. Sativa* was found to be the most effective intervention, where the meta-analysis of three studies showed that it reduces TG by −21.09 mg/dL (95% C.I.: −44.96 to −2.77; *p* value = 0.08). Although the confidence interval of the effect size is wide and the *p* value of the final effect is 0.08, the heterogeneity among the studies was fairly low at 28%. *Lashunadi guggulu*, according to the combined results of two small studies, reduces TG levels by 13.23 mg/dL (95% C.I.: −28.53 to 2.07; *p* value = 0.09). Six studies on guggulu, when meta-analyzed, showed that guggulu helps to reduce TG levels by 7.35 mg/dL (95% C.I.: −23.29 to 8.59; *p* value = 0.0.37). Here, two studies that showed positive results in other cholesterol levels are negative and opposite in one study.


**Table5.**EffectofAyurvedicherbalpreparationsontriglycerides(mg/dL).

**Figure 4.** Forest plot on effect of Ayurvedic herbal preparations in triglycerides (mg/dL).

#### 3.1.4. HDL (mg/dL)

The meta-analysis of 21 RCTs with 1186 participants (615 in the Ayurvedic group and 571 in the placebo group, as shown in Table 6) sugges<sup>t</sup> the statistically non-significant effect of Ayurvedic interventions on HDL-C. Guggulu alone and when mixed with garlic, however, showed positive and statistically significant results in increasing HDL-C. Analysis of end results from five RCTs in guggulu involving 264 total subjects showed that, as compared to the placebo, guggulu increased HDL-C by a small but significant difference of 2.19 mg/dL (95% C.I.: 0.27 to 4.12; *p* value = 0.03). On the other hand, results from a single study showed that *Lashunadi guggulu* was found to be raising HDL-C levels by 10 mg/dL (95% C.I.: 5.87 to 14.13; *p* < 0.00001). *N. sativa* also did not seem to have a significant effect on HDL-C. Results of the meta-analysis of three studies showed that it raised HDL-C levels by 1.92 mg/dL (95% C.I.: −1.62 to 5.45; *p* = 0.29). These studies involved a total of 163 participants. Garlic was also found to have no significant effect on HDL-C levels. Among 12 studies involving 736 participants, five studies claimed that garlic reduces HDL-C, and one study by Gardner et al. [51] found out that garlic neither reduces nor increases HDL-C. The studies were also highly heterogeneous with a high 97% I2 statistic. Though it may not even be relevant to conduct a meta-analysis on the effects of garlic on HDL-C, it is presented in the forest plot to show the current evidence (Figure 5).

**Table 6.** Effect of Ayurvedic herbal preparations on HDL-C (mg/dL).



**Figure 5.** Forest plot on effect of Ayurvedic herbal preparations in HDL-C (mg/dL).
